Homological Quantum Mechanics

We provide a formulation of quantum mechanics based on the cohomology of the Batalin-Vilkovisky (BV) algebra. Focusing on quantum-mechanical systems without gauge symmetry we introduce a homotopy retract from the chain complex of the harmonic oscillator to finite-dimensional phase space. This induces a homotopy transfer from the BV algebra to the algebra of functions on phase space. Quantum expectation values for a given operator or functional are computed by the function whose pullback gives a functional in the same cohomology class. This statement is proved in perturbation theory by relating the perturbation lemma to Wick's theorem. We test this method by computing two-point functions for the harmonic oscillator for position eigenstates and coherent states. Finally, we derive the Unruh effect, illustrating that these methods are applicable to quantum field theory.


Introduction
There are two fundamental formulations of quantum mechanics (and, by extension, of quantum field theory).There is the canonical formulation based on states as vectors of a Hilbert space and physical observables as self-adjoint operators acting on this space.This formalism determines amplitudes from which one then computes probabilities for physical observations.Arguably this is still the definite formulation of quantum mechanics.Importantly, however, there is also the Feynman path integral formulation in which the probability amplitude between states |q i ; t i and q f ; t f | is represented as Here S is the classical action for the dynamical variable q(t), and the integral is to be taken over the space of all trajectories (with fixed initial value q i and final value q f ).The path integral formulation is advantageous in many respects: First, starting from the classical theory encoded in the action S, the path integral directly provides the objects of physical interest, the probability amplitudes, without having to deal with Hilbert spaces, states, operators, etc.Second, any symmetries realized as invariances of the action S, such as spacetime Lorentz invariance, are manifestly realized in the quantum theory (in the absence of anomalies).In fact, much of modern quantum field theory would be unthinkable without Feynman's path integral formulation.The trouble is that it has turned out to be very difficult, if not impossible, to give a mathematically rigorous definition of the integral over the space of all kinematically allowed trajectories.One may discretize the physical problem at hand, in which case the path integral reduces, after Wick rotation, to well-defined finite-dimensional Gaussian integrals.This has been applied with great success in lattice gauge theory, but a general and non-perturbative definition of Feynman's path integral remains out of reach.
In this paper we introduce an alternative algebraic formulation of quantum mechanics based on the (co-)homology of the Batalin-Vilkovisky (BV) algebra [1,2].In this we pick up on a point that to the best of our knowledge was first made by Gwilliam [3] and by Gwilliam and Johnson-Freyd [4] for finite-dimensional toy models.We generalize their formulation to be applicable to genuine physical theories.Like the path integral formulation, this approach has the advantage of providing a direct way to pass from the classical theory to the physical quantities of the quantum theory, without having to invoke Hilbert spaces and the like.However, unlike Feynman's path integral, whose rigorous definition would require an infinite-dimensional generalization of calculus, the formulation presented here is algebraic, employing in particular methods from algebraic topology.It should be emphasized from the outset that we do not claim to provide a full-fledged alternative formulation of quantum mechanics.Rather, at the present moment, this formulation is restricted to the computation of normalized quantum expectation values with respect to a certain class of states.
In order to begin explaining this homological approach let us consider a one-dimensional dynamical system with dynamical variable q(t).The space of all kinematically allowed trajectories is then the infinite-dimensional vector space C ∞ (R) of smooth functions of one real variable.This space being infinite-dimensional does not cause any trouble in classical physics, where the equations of motion effectively make the problem finite-dimensional: a solution is uniquely determined after picking initial or boundary conditions, say by fixing the two values q(t i ) and q(t i ) at some initial time t i .As such, in classical physics the real space of interest is just R 2 , which can be viewed as the phase space of this system.In contrast, in quantum mechanics it is not sufficient to go 'on-shell': the path integral should be taken over all functions q(t) in C ∞ (R), with the ones solving the classical equations of motion being only the dominant contribution in the limit of small .At this stage concepts from topology turn out to be helpful, notably the notion of homotopy.One considers two shapes or spaces to be homotopy equivalent if one can be smoothly transformed into the other, as for instance a closed curve which, without tearing it, can be contracted to a point.Thus, a one-dimensional curve may be homotopy equivalent to a zero-dimensional point.Similarly, we will show that passing from the infinite-dimensional space C ∞ (R) of kine-matically allowed trajectories to the finite-dimensional phase space R 2 can be viewed as what is called a homotopy retract.While under homotopy one of course looses some information, the important fact is that the (co-)homology is homotopy invariant, this being one of the core techniques in algebraic topology.Since, as will be argued here, the cohomology encodes the objects of physical interest in quantum mechanics this technique allows us to circumvent the hard problem of making sense of the path integral over the infinite-dimensional space of all trajectories and to give a more direct homological definition of quantum expectation values.
In order to discuss our main technical results in some detail, we begin with the algebraic structures encoding the data of physical theories.A perturbative classical field theory is most directly encoded in a Lie-infinity or L ∞ algebra.This is a differential graded generalization of a Lie algebra, i.e., a graded vector space V equipped with a differential ∂ obeying ∂ 2 = 0 and a potentially infinite number of higher brackets or maps obeying generalized Jacobi identities [7], see [8][9][10][11][12] for reviews.The subspace V 0 of degree zero is the 'space of fields' and the subspace V1 of degree one is the space in which the equations of motion live or, in BV language, the space of anti-fields (in the example above both spaces are C ∞ (R)).Spaces of higher and lower degrees then encode gauge symmetries, gauge for gauge symmetries, Noether identities, etc. [12].(See also [13][14][15] for the L ∞ formulation of effective field theory in terms of homotopy transfer.)Given that ∂ 2 = 0 we can consider the cohomology of V : the space of ∂-closed vectors modulo ∂-exact vectors.The cohomology encodes the on-shell data, i.e. the (perturbative) solutions of the classical equations of motion modulo gauge transformations.
We can now explain the notion of a homotopy retract [16] (see also [17] for a pedagogical introduction to the closely related notion of homotopy transfer).One defines a projector from the infinite-dimensional space of fields to the finite-dimensional phase space, for instance by projecting a field φ to its initial and final values: p(φ) = (φ(t i ), φ(t f )) . (

1.2)
There is also an inclusion map i : R 2 → V 0 that reconstructs the solution from the initial and final values of φ.We have a homotopy retract if there is a homotopy map h : V 1 → V 0 from anti-fields (or equations of motion) to fields so that The first relation holds by definition of p and i, and we will show that the second relation holds upon defining h in terms of the Green's function.These relations tell us that while p has a 'right-inverse' it does not have a 'left-inverse'; rather, i is only a left-inverse 'up to homotopy'.This is as it should be since p is a genuine projector onto the much smaller space R 2 .This space in fact equals the cohomology of V , which is homotopy invariant.
In order to use this homotopy retract for quantum mechanics we then pass over to the closely related BV algebra which is defined on the 'dual' space of smooth functions on V (or rather functionals since V is typically infinite-dimensional), which we denote by F(V ). 1 The algebra structure is given by the graded commutative and associative product of functions together with a map ∆, called the BV Laplacian, which obeys ∆ 2 = 0. Importantly, ∆ is not a derivation of the product but rather is a differential of 'second order'.This implies that the failure of ∆ to act as a derivation on the product defines a new structure: the anti-bracket {•, •} on functions, which is a graded Lie bracket.In the conventional BV formalism one defines, given an action S, the homological vector field Q = {S, • }, which acts as a derivation on functions and obeys Q 2 = 0. From this one defines the BV differential which satisfies δ 2 = 0, provided the action satisfies the Maurer-Cartan or master equation −i ∆S + 1 2 {S, S} = 0.One also introduces an odd symplectic form from which the anti-bracket is derived like the Poisson bracket in standard symplectic geometry.The homological vector field Q is then the Hamiltonian vector field for the 'function' S, and the symplectic form is Qinvariant.The BV algebra employed in this paper deviates from the standard construction in the following subtle way: The BV differential is defined by (1.4), but Q is not the Hamiltonian vector field for S but rather equals {S, • } only up to boundary terms.The symplectic form would then not be invariant under Q, but in our construction this is immaterial since the symplectic form makes no appearance.Such generalizations of the BV formalism were introduced by Cattaneo, Mnev and Reshetikhin in [19,20].
The homotopy retract from V to R 2 gives rise to a homotopy retract from the BV algebra on F(V ) to the space of functions F(R 2 ).The BV algebra is thus transferred to the ordinary algebra of functions on R 2 concentrated in degree zero, with no non-trivial differential left, but this still encodes the complete cohomology of δ.Our core technical claim is now that the functions on R 2 in the cohomology space compute quantum expectation values as follows: Given a functional F of fields whose quantum expectation value we want to compute (for instance, for a 2-point function one considers F = φ(t)φ(s) for fixed times t, s) one determines the functional F ′ that is equal to F in cohomology (so that F ′ − F = δG for a suitable G) and that is just the pullback of a function f on R 2 , i.e., F ′ = f • p, where p is the projector in (1.2).This function f computes the following normalized quantum expectation value: where T denotes the time ordering operator, and |x; t is the state satisfying φ(t)|x; t = x|x; t , and similarly for y.More generally, we introduce a homological procedure to compute such normalized expectation values with respect to states that are any linear combination of position and momentum eigenstates.
In perturbation theory, one can give an effective procedure to determine the function f using the so-called perturbation lemma of homological algebra.In this case one can prove the above claim by showing that the computations are equivalent to the familiar techniques based on Wick's theorem in quantum perturbation theory.Our formulation is also related to other approaches in the literature on the homotopy algebra formulation of perturbation theory, see [21][22][23][24][25].While some technical details are similar, and in particular ref.[22] has been very useful for us, the general homological formulation presented in this paper turns out to be quite different, as we will discuss in more detail in the summary section.We will also give a formal path integral argument supporting the above claim.Most intriguingly, the homological formulation of quantum mechanics is not restricted to perturbation theory, yet potentially is mathematically well-defined.
The rest of this paper is organized as follows.In sec. 2 we introduce the needed background material on BV algebras, and present the main statement of the homological formulation of quantum mechanics.We also set the stage for our subsequent applications by giving the homotopy retract for the harmonic oscillator.In sec. 3 we compare the homological formulation with standard quantum mechanics in perturbation theory and using Feynman's path integral.We then illustrate and apply the homological approach in sec.4 by computing expectation values for the harmonic oscillator with respect to position eigenstates and coherent states, and we verify that the results agree with those of conventional quantum mechanics.In sec.5 we show that this approach is applicable to quantum field theory by re-deriving the Unruh effect, which to the best of our knowledge provides a new derivation of this effect.We close in sec.6 with a summary and outlook.In order to keep the paper self-contained we include an appendix summarizing the key concepts from homological algebra.

General Approach
The goal of this section is to give the homological formulation of quantum mechanics outlined above.In the first subsection we define the BV algebra and review the finite-dimensional case following Gwilliam and Johnson-Freyd [3,4].We then turn in the second subsection to genuine quantum-mechanical systems and state the main claim about the homological computation of quantum expectations values.In the final subsection we introduce a homotopy retract for the harmonic oscillator, as preparation for the applications in later sections.

BV algebra
We begin by reviewing the BV algebra and the homological formulation for a finite-dimensional toy model, following sec.3 in [3].Thinking of an 'action' function S(x) of a finite number of variables x i , i = 1, . . ., N , the BV formalism is defined on the larger space of functions F (x, x * ) of x i and new anti-commuting variables x * i .We typically consider functions of the form where the coefficients are smooth functions of x (for instance polynomials) and antisymmetric in i 1 . . .i k .This expansion gives a grading to the vector space of functions according to the number of x * : functions depending only on x are of degree zero, functions linear in x * are of degree −1, etc.A useful mnemonic is to assign a ghost degree of zero to x and a ghost degree of −1 to x * .In the following we will display algebraic relations for homogenous functions of fixed degrees, denoting the degree of F by |F |, etc., and we also write (−1) F ≡ (−1) |F | .More general relations then follow by linearity.The multiplication of functions equips this space with a graded algebra structure with F Since the variables x * i anti-commute one has to be careful with the notion of derivative: there are left-and right-derivatives which are defined operationally by the first-order variations We follow the convention that, unless explicitly indicated otherwise, all derivatives are left derivatives.Also note that second-order derivatives w.r.t.
. These derivatives also obey graded Leibniz rules w.r.t. the multiplication of functions: In the next step one equips this space with a graded Lie bracket and a differential.The Lie bracket is called the anti-bracket and defined like the Poisson bracket: This bracket has intrinsic degree of +1, i.e. |{F, G}| = |F |+|G|+1, and is graded antisymmetric with the degrees shifted by one, i.e., The anti-bracket obeys the graded Jacobi identity and the following compatibility condition with the product of functions: There is a differential of intrinsic degree +1, called the BV Laplacian, defined as (2.8) The BV Laplacian squares to zero, which is a consequence of ordinary derivatives commuting while derivatives w.r.t.x * anticommute.Furthermore, the Laplacian acts as a derivation on the anti-bracket: ∆{F, G} = {∆F, G} + (−1) F +1 {F, ∆G} . (2.10) In order to prove this it is convenient to first note that ∆, being a second-order differential operator, does not act as a derivation w.r.t. the usual multiplication of functions, but rather the anti-bracket encodes the failure of ∆ to act as a derivation: This follows quickly from the definitions (2.4) and (2.8).Acting on this relation with ∆, using ∆ 2 = 0 and this relation again, then establishes the Leibniz relation (2.10).This state of affairs can be summarized by saying that the anti-bracket {•, •} together with the BV Laplacian ∆ form a differential graded Lie algebra (or an L ∞ -algebra without higher brackets).More precisely, here the grading of the differential graded Lie algebra is given by the above grading shifted by one, explaining the presence of factors like F + 1 in many formulas.
Let us pause here to define the notion of a BV algebra.To motivate this definition we need to establish another relation.Acting with the Laplacian (2.8) on the product of three functions yields by a straightforward computation using (2.3) (2.12) Upon using (2.11) this can be rewritten in a form that employs only the (graded) multiplication of functions and the BV Laplacian: (2.13) We can now define a BV algebra:

Definition:
A BV algebra is a vector space equipped with a graded commutative and associative product and a degree-(+1) map ∆ that satisfies ∆ 2 = 0 and is of second order in the sense of (2.13).
Note that the above definition only refers to the product structure and the differential ∆, which is the only unconventional ingredient since it does not obey the Leibniz rule with respect to the product but rather the 'second-order Leibniz rule' (2.13).The differential graded Lie algebra structure for the anti-bracket is then a derived notion, with the anti-bracket being defined by (2.11).All relations of the differential graded Lie algebra then follow from the axioms of a BV algebra as does the compatibility relation (2.7), which can be derived from (2.11) and (2.13).

Master Equation and Cohomology
Returning to the application of BV algebras in physical theories we note that for any differential graded Lie algebra one can write the Maurer-Cartan equation for a vector (function) of degree one (which in the above grading is degree zero).An example of such a function is the action S, and the Maurer-Cartan equation reads where the exponential of the degree-zero object S is defined in terms of its Taylor series.This relation can be verified with (2.11).
Since δ 2 = 0 there is a notion of homology spaces ker δ im δ : spaces of δ-closed vectors modulo δ-exact vectors.Our goal is now to compute this homology for a simple toy model with one real variable x and action (2.18) We will see that the homology computes the 'path integral' including e i S , which here reduce to regular integrals.
We begin by writing out the differential δ defined in (2.15), using that here only two variables x and x * with (x * ) 2 = 0 enter: First, we have to determine ker δ, the space of functions that are δ-closed.A general function can be written as the superfield where as above we assume that f and g are polynomials.This yields Setting this to zero gives a differential equation that does not have a non-trivial solution in polynomials (otherwise one has the solution e ia x 2 2 ).Therefore, the kernel of δ is given by functions (2.20) with g = 0: Next we have to mod out by im δ, the space of δ-exact functions.In order to see the significance of this step let us consider functions of the form (2.20) with f = 0 and g = x n , which yields with (2.21) (2.23) 2 Using (2.7) one may verify that Q ≡ {S, •} acts as a derivation with respect to the product, so that with (2.11) the anti-bracket can also be defined in terms of the full BV differential as When passing over to homology we identify any two functions that differ by a δ-exact function.
Hence the above relation implies that its right-hand side is zero in homology or, equivalently, that we have the equivalence For instance, x 2 is equivalent (can be reduced to) a constant, x 2 ∼ i a .Similarly, x 3 can be reduced to x, since x 3 ∼ i a 2x, but noting that x is δ-exact, x = − 1 a δ(x * ), this is actually equivalent to zero.Using this iteratively we have The reader may recognize the right-hand side as the 'quantum expectation value' x n of a zero-dimensional QFT with a single variable (see, e.g., p. 14 in Zee's text book [27]).More precisely, let us define the expectation value for a polynomial f in terms of the convergent Gaussian integral as ) , where the substitution a → −ia is done after performing the integral.Then the right-hand side of (2.25) equals x n .More generally, any polynomial f is equivalent in homology to a complex number, and identifying its homology class [f ] with this number we have Thus, the homology of the differential δ computes expectation values.

BV for Quantum Mechanics
We now turn to generic dynamical systems and state our main claim about the homological formulation of the corresponding quantum theories.For definiteness let us consider a onedimensional mechanical model with action (2.28) The fields here are smooth functions φ on the interval [t i , t f ], i.e. φ ∈ C ∞ ([t i , t f ]).Formally, such theories include genuine (quantum) field theories, where the dependence of φ on spatial coordinates is suppressed, but for definiteness let us think of ordinary quantum mechanics.
We start from the BV formalism as in the previous section.This means that the space of dynamical variables is enlarged to include, in addition to φ, anti-fields φ * , which we also assume to be smooth, φ * ∈ C ∞ ([t i , t f ]).Fields and anti-fields are combined into a total space which we also sometimes denote by V • , to indicate that we refer to the total space of a chain complex.Moreover, here we used, for any vector space X, the common notation ΠX to indicate that the elements of ΠX have reversed parity compared to X.This means, for instance, that taking the elements φ ∈ C ∞ ([t i , t f ]) to be of even degree, the elements φ * ∈ ΠC ∞ ([t i , t f ]) have odd degree, although in the end they are both just smooth functions.The definition of the BV algebra given in the previous subsection goes through in the present context, just with functions replaced by functionals and derivatives replaced by functional derivatives.Concretely, we consider functionals that are superpositions of the monomials where the coefficient functions f (t 1 , ..., t k , s 1 , ..., s l ) are completely symmetric in the t i and completely antisymmetric in the s i .The degree of such a functional is minus the number of anti-fields appearing in there, i.e., |F | = −l.(The relative minus sign compared to the degree of +1 for φ * encoded in (2.29) is not a typo but rather due to functionals being dual to (anti-)fields, hence having the opposite degree.)We denote this space of functionals by F(V ), which inherits a grading by the number of φ * , so that we can write The BV algebra structure is defined on this space as in the previous subsection.For instance, the differential δ = Q − i ∆, which still satisfies δ 2 = 0, consists of two parts.There is the classical piece (surviving the limit → 0) given by the homological vector field where EL(φ(t)) = 0 are the Euler Lagrange equations.It also satisfies . (2.32) Note that Q and δ both decrease the number of φ * by one and hence have an intrinsic degree of +1.(Sometimes we denote by Q p the restriction of Q to F(V ) p .) We have to comment on the following deviation from the standard BV formalism: in general δS δφ(t) is not equal to EL(φ(t)), and so in general Q = {S, −}.The reason is that these two are only equal up to boundary terms.(Later it will be crucial to allow for variations along solutions, and these variations do not vanish on the boundary.)Indeed, we take the space of fields to be the space of smooth functions defined on the interval [t i , t f ] rather than the entire real line.This makes the action well-defined without having to assume any behavior of the fields at infinity.However, in contrast to standard treatments of the BV formalism, the symplectic form ω inducing the bracket {−, −} is no longer invariant under the vector field Q due to boundary terms.An extension of BV formalism, called BV-BFV formalism, accommodates such features and was developed by Cattaneo, Mnev and Reshetikhin in [19,20], but in this work we never use the symplectic form, so these issues are of no concern to us.
Given that δ 2 = 0 we can define the cohomology of F(V ) with respect to δ, and we will show that this cohomology computes quantum expectation values.Before turning to this it is instructive to inspect the cohomology 0. Specifically, we will first establish the following Claim: The cohomology of Q is isomorphic to the space of "on-shell functionals".
Here on-shell functionals refer to functionals on solutions to the equation of motion EL(φ(t)) = 0.More precisely, this holds under the assumption that there are no non-trivial gauge symmetries.In the following we will establish this statement.To this end let us introduce some notation: We denote the subspace of solutions by E ⊆ V 0 , Since E is a subspace of V 0 , any functional F on V 0 can be restricted to a functional on E. This defines a projection Let us assume that any functional on E is obtained in this way.In other words, any functional on E extends to a functional on V 0 .Then the restriction map r is surjective.In general, however, r is not a bijection since it may have a non-trivial kernel, but upon modding out the kernel we have an isomorphism: We now compute the cohomologies in order to establish the above claim.We begin at degree zero and note that functionals of degree zero are annihilated by Q, hence ker Q 0 = F(V 0 ).We next show that in degree zero the image of Q consists of functionals of φ proportional to the equation of motion.To see this consider a functional of degree −1, i.e., G −1 = and so this functional vanishes on E. Therefore, Q(G −1 ) is in the kernel of r, i.e. im Q −1 ⊆ ker r.
Assuming that any functional in the kernel of r is of the form (2.35) (c.f.appendix B in [6] for a discussion on this assumption), for a suitable function g −1 , one has in fact im Q −1 = ker r.
We then find for the cohomology So the cohomology in degree zero is in fact equal to the space of functionals on solutions.This establishes the claim for functionals of degree zero.
Let us now turn to the cohomology in degree minus one.The kernel of Q −1 consists of functionals G −1 of the form given above (2.35) for which Q(G −1 ) = 0.There is a straightforward physical interpretation of this condition: inspecting the explicit form (2.35) one infers that is a (gauge) invariance of the action.We next investigate which functionals at degree −1 are in the image of Q −2 .Consider a generic functional at degree −2, (2.37) Computing Q(G −2 ) one obtains functionals of the form G −1 given above (2.35)where Under the interpretation of gauge symmetries these are precisely the trivial gauge symmetries, which always exist.Their action vanishes once we restrict to solutions.We therefore learned that the trivial symmetries are in the image of Q −2 and hence that ker Q −1 im Q −2 parametrizes nontrivial gauge symmetries.When there are no non-trivial gauge symmetries, the cohomology in degree −1 therefore vanishes.A similar argument applies in the case of the cohomologies in arbitrary negative degree.This completes our discussion of the claim that the cohomology encodes "on-shell functionals", i.e., functionals on solutions of the equations of motion.
For the following applications it will be important to have an explicit model for the space of solutions.Suppose that φ 0 is a solution to the equation of motion.Any such solution is uniquely determined by its boundary values x i = φ(t i ) and x f = φ(t f ).We therefore have E ∼ = R 2 and so F(E) ∼ = F(R 2 ).Any functional on the space of solutions can be viewed as an ordinary function f (x i , x f ) in two variables.
The space R 2 of boundary values naturally embeds into V • .Given (x i , x f ) ∈ R 2 , let φ x i ,x f be the unique solution with these boundary conditions, i.e., we demand that (2.39) This defines the embedding or inclusion map There is also an associated projection satisfying p • i = id R 2 according to (2.39).
Given the above maps we can define their pullbacks that act (in the opposite direction) on the dual spaces of functionals and functions.The pullback of the inclusion is the map Thus, this map associates to a functional F the function on R 2 defined by i Similarly, the pullback of the projection is the map We say that a functional F restricts to R 2 , or that F is an "on-shell functional", if it is the pullback of a function on R 2 .In order to identify ker Q im Q with R 2 , we can look for representatives is the pullback of a function f on R 2 and can be hence be written as As an aside we note that the pullback maps i * and p * define what are called chain maps in the homological language.When we think of F(R 2 ) as a complex in degree zero with trivial cohomological vector field Q = 0, the maps maps i * and p * obey The first relation follows since the image of i are the solutions on which Q vanishes, while the the second relation follows since f • p is a functional of degree 0.
We now turn to our main statement relating to quantum mechanics.Thinking of −i ∆ as a small perturbation, we expect that the cohomology of δ = Q − i ∆ is the same as that of Q.We can then search for functionals in the δ cohomology class of F that restrict to R 2 , i.e. we look for a functional G, such that Let us point out that while the above statement was made in the context of a particular projector mapping the infinite-dimensional vector space V to R 2 , this relation between cohomology and quantum expectation values holds more generally.For instance, we may consider the more general projectors (2.47) In this case, the boundary states change.The incoming state |x, t i is now an eigenstate of the operator a i φ(t i ) + b i π(t i ) with eigenvalue x, where π(t i ) is the momentum operator at t = t i .Similarly, the outgoing state y, t f | is an eigenstate of a f φ(t f ) + b f π(t f ) with eigenvalue y.

Homotopy Retract for Harmonic Oscillator
In the previous subsection we have argued that for a generic dynamical system the infinitedimensional function space of dynamical variables is equivalent in cohomology to the finitedimensional phase space (or the space of initial conditions).Here we discuss further the topological interpretation of this fact by displaying, for the case of the harmonic oscillator, a so-called homotopy retract that maps between the corresponding chain complexes.This will be instrumental for the applications to be discussed below.
The equation of motion for the harmonic oscillator is and defines the BV differential Q according to (2.31).Since the cohomology of Q describes the space of functionals on the solution space of φ + ω 2 φ = 0, which is isomorphic to the space of initial conditions (φ(t i ), φ(t i )) ∈ R 2 , we expect that the cohomology of Q should be isomorphic to functions on R 2 .Instead of working on the spaces of functionals, on which Q acts as a vector field, we can work directly on the field space encoded in the complex where we recall that 3 The cohomology of this complex in degree zero is ker ∂, so it consists of the space of solutions.Since any solution is determined by its initial condition (φ(t i ), φ(t i )), we can identify the cohomology with the phase space R 2 .In degree one, the cohomology is zero, because any element is ∂-exact: for any function f (t) there is a φ(t) so that From the theory of ordinary differential equations we know that a solution, which we denote by φ f , exists under very mild assumptions on f .Smoothness of f ensures that φ f is also smooth.We can give an explicit solution to (2.50) by means of a Green's function: Explicitly, the corresponding kernel function reads with θ the step function, and it satisfies Of course, φ f (t) is unique only up to homogeneous solutions, i.e. solutions to (2.48).We picked a solution such that φ f (t i ) = φf (t i ) = 0.
Our above analysis shows that the cohomology space ker ∂ Im ∂ is isomorphic to the phase space R 2 , which we can think of as a chain complex in degree zero with trivial differential.We have the following maps called quasi-isomorphisms: There is also a quasi-isomorphism i : R 2 → V 0 in the other direction, defined by In the following we will construct a so-called homotopy retract of the original chain complex V • to its cohomology R 2 .To this end we interpret the Green's function (2.51) as a map h : (2.57) We can extend h to the full complex V • by defining h(φ) = 0 for any field φ ∈ V 0 .The map h is called a homotopy retract from the original complex V • to its cohomology R 2 .In general, we say that a complex (Y, ∂ Y ) is a homotopy retract of (X, ∂ X ) if there are chain maps A direct computation with p and i defined in (2.55) and (2.56) shows that More specifically, the first relation is a direct consequence of the property (2.53), while the second relation requires two integrations by part in order to use that the Green's function obeys the analogue of (2.53) with respect to its second argument.In this computation, one picks up boundary terms that precisely constitute −ip(φ).We thus obtain the homotopy relation (2.58) if we define p and i to be zero on V 1 .This shows that the maps from V • to R 2 define a homotopy retract.
We should emphasize that the above choice of Green's function and homotopy is just one of many.The kernel K of the above retarded Green's function is, in particular, not symmetric under t ↔ s.We can also choose Dirichlet boundary conditions at t i and t f , for which the kernel K DD (t, s) is symmetric and given by This yields the homotopy map Although in the form (2.60) the symmetry is not manifest, a non-trivial computation shows K DD (t, s) = K DD (s, t). 4 The corresponding chain maps are

.64)
As a next step we lift the homotopy retract from the underlying spaces V • and R 2 to the BV complex, i.e., to the space of functionals on V • and the space of functions on R 2 .We recall that the differential on this complex is given by the BV differential, which reads for the harmonic oscillator . (2.65) In order to realize the homotopy retract on the BV complex we have to use that, as explained above, the maps i and p can be applied to functionals via pullback to give I := p * and P := i * .Specifically, for a functional F [φ, φ * ] in F(V • ) we obtain the function on R 2 : while for a function f ∈ F(R 2 ) we obtain the functional on (2.67) We next have to show how to use the homotopy h on V • to define a homotopy on F(V ).
We first consider the linear functionals φ(t) and φ * (t).To clarify this notation let us point out that the symbol 'φ(t)' can be interpreted in two natural ways.The standard interpretation is that of a function φ that takes different values depending on the variable t: here we think of φ as fixed and of t as a variable.The second interpretation of 'φ(t)' takes t to be fixed but φ to be variable.This then defines a functional: a map that assigns a number to each function φ, namely the number φ(t) obtained by evaluating the function φ at the fixed t.Similar remarks apply to the interpretation of 'φ * (t)' as a function or as a functional.A subtlety of this notation is that the degree of φ * (t) depends on its interpretation: as a function it has degree +1, but as a functional it has degree −1, c.f. the discussion after (2.30) above.It will always be clear from the context which interpretation applies.
Returning to the problem of defining a homotopy map H on functionals we first define H on the linear functionals φ(t) and φ * (t) by (2.68) Note that H(φ(t)) does not actually depend on φ(t).The functional H(φ(t)) maps a given anti-field φ * to the number that is given by the integral on the right-hand side.This functional is linear in φ * and so has degree −1, as it should be since H has intrinsic degree −1.It is now straightforward to show that (2.69) To obtain a homotopy on all functionals, we observe that any polynomial functional is a superposition of products of the functionals φ(t) and φ * (t).So it is enough to know how the homotopy distributes over any product of functionals.We set Note that we did a graded symmetrization over F and G on the right-hand side, which is necessary since H(F G) should be graded symmetric in F and G.The definition of H ensures that the homotopy relation is satisfied on products F G, as long as it holds on F and G individually.The action of H on any functional F can then by successively reduced to its action on φ(t) and φ * (t), where we define it via (2.68).This shows that p * i * is homotopic to the identity.So F(V • ) is homotopic to F(R 2 ).We therefore proved that the space of functionals on V • is quasi-isomorphic to the space of functions on phase space R 2 .

Comparison with Standard Formulations
In this section we verify from various angles the main statement (2.46) that relates the cohomology of the BV differential to quantum expectation values.First, we explain the perturbation lemma and show that it relates to Wick contractions in the familiar formulation of quantum mechanics.Second, we give a heuristic argument based on the path integral formulation.

Perturbation Lemma
The homological perturbation lemma provides a recipe to compute the cohomology of the BV differential and in particular the representative F ′ appearing in the main statement around (2.46).The perturbation lemma states the following.Suppose that we have homotopic complexes (X, d X ) and (Y, d Y ).This means that there are quasi-isomorphisms p : (X, d X ) → (Y, d Y ) and i : (Y, d Y ) → (X, d X ), together with a homotopy h : (X, d X ) → (X, d X ), such that Now assume that we perturb the differential of X by some η, such that we still have a chain complex, i.e. d ′ . This means that there are new quasi-isomorphisms The perturbation lemma provides explicit formulas for the perturbed data: Here the perturbed differential d ′ Y is such that the perturbed projection and inclusion are chain maps, which means that If the complexes have additional structure, we want the perturbations to preserve these.For instance, the space of functionals forms an algebra under multiplication, and the cohomological vector field Q = {S, −} acts as a derivation.In order to preserve these properties we first have to assume that the original homotopy data preserves these.So i and p have to be algebra morphisms, i.e. i(F G) = i(F )i(G) and p(F G) = p(F )p(G).In addition, we have to assume that h is a strong deformation retract, which means that h is subject to the side conditions If these conditions are met, the perturbed chain maps p ′ and i ′ will again be algebra morphisms, and d ′ Y will be a derivation.We may first apply the homological perturbation lemma to a classical mechanical system.Given an action S, we split it as S 0 + S I , where S 0 is a free theory (quadratic in fields) and S I is the interaction term.We now think of (F(V ), Q 0 = {S 0 , −}) as our initial complex, which is based on the free action (a sum of harmonic oscillators).We saw that a Green's function defines a homotopy on the complex (V • , ∂) to phase space, satisfying {∂, h} = 1 − i • p, which in turn defines a homotopy H from F(V ) to F(R 2 ), c.f. (2.70), satisfying Such a homotopy in terms of a Green's function defines a strong deformation retract, i.e. the analogue of (3.5) holds for H.We then take the interaction term Q I = {S I , −} as the perturbation.Since the functions on solution space are concentrated on degree zero, there will be no induced differential, but the space (F(R 2 ), 0) is still a trivial complex.Since Q 0 + Q I encodes the full interacting equations of motion P ′ projects functionals to solutions of the full equations of motion.So this is the role of the projector P ′ : it perturbatively constructs solutions of the interacting theory using the Green's function and a given solution of the free theory.
We now turn to our core application for the quantum case.The perturbation lemma allows us to show that in perturbation theory any functional F of degree zero has a representative F ′ in the cohomology of δ = {S, −} − i ∆ that can be written as F ′ = f • p = I(f ) for some function f on phase space.We use a Green's function to define a homotopy and view η = {S I , −} − i ∆ as a perturbation.With the perturbation formulas (3.3) we obtain the new chain maps where the second equation is true by degree reasons. 5We then define The claim is that these are the desired functional and function.Indeed, F ′ is in the same cohomology class as F , since where we used the homotopy relation together with H(δ(F )) = 0 for degree reasons.So the objects required by the statement around (2.46) are given by

Wick Contractions
The perturbation lemma discussed above gives a complete perturbative solution to the problem of determining the functional F ′ that in cohomology is equivalent to F , together with the function f .However, the required action of the homotopy map H given in (2.70) is inconvenient for computations due to the explicit symmetrization.The action of the homotopy can be brought to a more manageable form by splitting the field space into off-shell and on-shell fields.This has the additional advantage of relating the perturbation lemma to standard computations in quantum field theory using Wick contractions.More precisely, for the special case that the homotopy is given by the Feynman propagator, the perturbation lemma amounts to Wick's theorem, but the perturbation lemma is more general, as we will see in the next section.
We begin by decomposing the field space by using the projection p and inclusion i: where V p is the 'physical' subspace and V u is the 'unphysical' subspace.Any field can now be written as φ = φ p + φ u , where φ p is a solution to the equations of motion.Note that φ u depends on the choice of homotopy and satisfies the same boundary conditions as the Green's function used to construct the homotopy.In the dual picture of functionals on V we have the induced decomposition where F p = F(V p ) is the space of functionals depending on φ p only.More precisely, these are the functionals of the form (2.30) where all φ are φ p and there are no φ * .Correspondingly, F u is the space of functionals that contain at least one φ u or at least one φ * .
Next we consider the above decomposition at the level of the (functional) derivatives defining the BV algebra.Recall that the space of physical or on-shell fields is isomorphic to a finitedimensional space (in the present case V p ∼ = R 2 ), so that we can use ordinary coordinates (x, y) of R 2 to label a solution φ p;x,y .This allows us to introduce partial derivatives along solutions: We also introduce a functional derivative in the direction of where g u ∈ V u .The formula for δF [φ,φ * ] δφu(t) is the same as for the functional derivative δF [φ,φ * ] δφ(t) , the only difference being that the space of functions entering the integral is restricted.For φ * we do not introduce a new notation, since the φ * are unaffected by the decomposition.
We next give a chain rule relating δ δφ(t) to ∂ x , ∂ y , δ δφu(t) .Indeed, we can represent a functional F [φ] as a functional F [φ u , x, y] (momentarily suppressing the dependence on φ * ) by inverting the relation φ shows that also the boundary terms vanish. 6The resulting vector field has been called N , because it counts the total number of fields φ u and anti-fields φ * : We claim that on the space F u of functionals that are at least linear in φ u or φ * , on which N is a positive operator, the homotopy map is implemented by where we have identified N with its eigenvalue (which is always positive on F u , so N −1 is well-defined).Indeed, the homotopy relation is then satisfied: recalling that the subspace F u is projected to 0 (or, equivalently, the homotopy H u is a strong deformation retract of F u to 0).Finally, we can extend H u to a homotopy H on the total space F(V ) = F u ⊕ F p by declaring H to be zero on F p ⊆ F(V ).We then have {Q 0 , H} = 1 − IP , which defines a strong deformation retract from F(V ) to F p .
Having defined, using the above decomposition, the lift of the homotopy map H we can now revisit the application of the perturbation lemma and relate it to Wick contractions.In the first step we consider the free harmonic oscillator with differential Q 0 = {S 0 , −} and view −i ∆ as the perturbation.Under this perturbation there will neither be a induced differential on R 2 , nor a perturbation to the inclusion I, while the perturbed projection is where we denote the perturbed projector by P 1 since below we will consider a second perturbation to be denoted P 2 .Since P 1 is of degree zero it is zero on functionals with at least one anti-field.On functionals of fields only, ∆H acts as 6 This can be easily checked for specific boundary conditions such as Dirichlet, but one can also give a general argument valid for arbitrary boundary conditions as follows.We write the Lie bracket as where B encodes possible boundary terms.Since N and {Q0, V h } are vector fields, so is B. We will show that B = 0. Since in (3.20) we saw that on the linear functionals φ = φu + φp and φ * (t), V h agrees with the homotopy H, we have {Q0, V h } = 1 − IP on these functionals.Since IP (φu(t)) = IP (φ * (t)) = 0, we find Comparing this with (3.23) acting on φu and φ * , for which N acts as the identity, we learn B(φu(t)) = 0 and B(φ * (t)) = 0, i.e., B is zero on these linear functionals.General functionals on Fu are superpositions of functionals of the form f (x, y)F [φu, φ * ], where (x, y) are coordinates on Vp and F [φu, φ * ] is at least linear in φu and φ * .The vector fields V h and Q0 act directly on F [φu, φ * ], ignoring the function f (x, y) in front, and so does {Q0, V h }.Since B(φu(t)) = B(φ * (t)) = 0 and B acts as a vector field via the Leibniz rule we have B = 0.
Recall that the action of δ δφ(t) reduces to δ δφu(t) when it is integrated against a function satisfying the boundary conditions of V u .The kernel K(t, s) is such a function and so defining we have with (3.29) the relation We now consider a functional F for fixed t 1 , . . ., t m of the form where f (x, y) is any polynomial in x and y.Any other functional is a superposition of these functionals, so the effect of P 1 on all functionals can be deduced from its effect on F by linearity.Since P (φ u (t)) = 0, the only non-zero contribution to P 1 in (3.28) comes from the term where n≥0 (i ∆H) n eliminates all fields φ u .This can only happen when m is even, i.e., m = 2k, for which only the term with n = k contributes.We then find with (3.31) This implies that on arbitrary functionals we have Note that to apply this formula, we do not need the split V p ⊕ V u , since because K(t, s) is symmetric.(If K(t, s) were not symmetric, it would not satisfy the boundary conditions of V u in the variable s.) The above results allow us to establish the core relation with the familiar techniques of perturbative quantum field theory: The operator C (which is called 2∂ P in lemma 3.4.1 in [5]) generates Wick contractions, and P 1 implements Wick's theorem upon interpreting P as a normal ordering operation.Of course, P depends on the choice of homotopy H, and therefore on the choice of propagator.We will see that, when choosing the Feynman propagator, P can indeed be interpreted as the usual prescription to move "annihilation operators to the right".So far we have applied the perturbation lemma to the free theory with differential Q 0 , viewing −i ∆ as the perturbation, and related this to familiar Wick contractions.Next we include the non-linear interactions of the classical theory, thereby combining the perturbations Q I = {S I , −} and −i ∆.Thus, we apply the perturbation lemma by taking Q I to be a perturbation of Q 0 − i ∆, for which one obtains the perturbed map (3.36) The core result following from the perturbation lemma is then that the function f computing the quantum expectation value for the functional F is given by f = P 2 (F ) . (3.37) We will now show that this prescription coincides with conventional quantum perturbation theory, i.e., that the operator P 2 computes expectation values of the full interacting theory.This is established if we can show that P 2 is in fact equal to P2 defined by with normalization because this is the familiar method: performing Wick contractions of the operator F weighted by exp( i S I ) with the interacting action S I .
The proof that (3.38) indeed represents P 2 follows [22], which we repeat in our setting.We first show that (3.38) is consistent with the relation P 2 • I = 1 that P 2 obeys, i.e., that (3.40) First, using HI = 0, we have n≥0 (i ∆H) n I(f ) = I(f ) and therefore with (3.28) = P I(f )P n≥0 (i ∆H) n e i S I = P I(f )Z = f Z . (3.41) Here we used that P I = 1 and the fact that P is an algebra morphism, i.e.P (F G) = P (F )P (G).We next use that the perturbed data with projection P 2 still obey the homotopy relation: where We apply P2 to both sides and use that P2 I = 1, c.f. (3.40).Then, Since P2 is non-zero only in degree zero, we find that P2 H 2 δ = 0, so (3.43) reduces to

.44)
We now show that Z P2 δ = 0, which then proves that P2 = P 2 .To this end we decompose the BV differential as δ = δ 0 + Q I .We assume that S I does not contain derivatives.We can then write Q I = {S I , •} due to the absence of boundary terms.We compute, for a generic functional F , by use of (2.16) (3.45) Here we used δ 0 e i S I = 0, which follows because both S 0 and S I contain no anti-fields, and {e X , F } = e X {X, F } for any degree-zero object X, which follows from (2.7).We then have Z P2 (δF ) = P 1 δF e i S I = P 1 δ 0 e i S I F = 0 , (3.46) where we used that P 1 δ 0 = 0, i.e. that P 1 is a chain map with respect to Q 0 − i ∆, as implied by the perturbation lemma.This concludes the proof of P 2 = P2 .
We end this part by summarizing what we have found.First of all, we showed that in perturbation theory, any functional F has a representative of the form F ′ = f • p.We showed that the perturbative computation of the Q 0 − i ∆ cohomology does Wick contractions, just like we would expect in a free theory.Also, we showed that for an interacting theory we can think of the homological perturbation lemma computing expectation values with respect to the free theory, but with functionals F weighted by the interacting part e i S I .This is the usual way in which Feynman diagrams are computed.

Path Integral
In the previous subsection we saw that Wick's theorem arises as a consequence of the perturbation lemma, proving that the latter entails in particular the usual treatment of quantum theories at the perturbative level.Nevertheless, we want to shed more light on the homological approach by comparing it to the path integral derivation.
In the path integral formulation, we formally compute expectation values of functionals by writing When working in the operator formalism, we have to include time ordering indicated by the letter T.However, we will often not write time ordering explicitly unless we want to stress its presence.
The integral on the right-hand side of (3.47) is thought of as being performed over all paths with φ(t i ) = x and φ(t f ) = y.We take the perturbative route and write S = S 0 + S I , where When boundary conditions are fixed, the operator ∂ 2 + ω 2 in S 0 is invertible, and the integral can be given a meaning in perturbation theory.
One way to proceed is to pick reference boundary conditions, e.g.φ(t i ) = φ(t f ) = 0. We then write φ = φ u + φ p , such that φ p is the unique classical solution with the generic boundary conditions φ p (t i ) = x, φ p (t f ) = y, while φ u satisfies φ u (t i ) = φ u (t f ) = 0. Explicitly, φ p is given by φ p = i(x(t i ), y(t f )) and can be viewed as a constant in the context of the path integral.We then substitute φ = φ u + φ p and assume that the integral measure is invariant under constant shifts so that Dφ = Dφ u .One computes Importantly, the boundary action ∂S does not depend on φ u , which is due to our choice of reference boundary conditions φ u (t i ) = φ u (t f ) = 0. Thus, the phase e i∂S(φp) can be scaled out of the path integral:

.51)
Picking F = 1 we have Our next goal is to define a map P 1 : F(V 0 ) → F(R 2 ) in terms of the path integral that equals the map in (3.34) that implemented Wick's theorem.We claim that this map can be written, for any functional F , as Indeed, once we accept that the path integral is computed by doing Wick contractions using the propagator with the boundary conditions specified on the right hand side of (3.53) it is clear that this is equal to the map (3.34).Note that P 1 is a function on phase space R 2 parametrized by the boundary conditions put on φ p .
We can finally give a path integral expression for the normalized quantum expectation value of F that agrees with our above result from the perturbation lemma.In terms of the map (3.53) we find with (3.51)  using that the phase e i∂S(φp) cancels.This is indeed the same as the map (3.38) defined by means of the perturbation lemma.This confirms that the conventional interpretation of the (otherwise ill-defined) path integral in perturbation theory agrees with the homological formulation.

Harmonic Oscillator
In this section we illustrate the homological formulation of quantum mechanics by applying it to the one-dimensional harmonic oscillator.Specifically, we compute two-point functions both with respect to positions eigenstates and with respect to coherent states.Using the familiar formulation of quantum mechanics, based on Hilbert spaces and operators, we then verify that the homological approach yields the correct results.We hope to illustrate in this that for certain computations, as in the case of position eigenstates, the homological approach is more transparent.

Homological Computation for Position Eigenstates
For a free theory like the harmonic oscillator all expectation values can be computed in terms of the two-point function via Wick's theorem, which we also obtained from the perturbation lemma.For this reason, we apply the homological formulation to the two-point function of the harmonic oscillator.The BV differential for the quantum harmonic oscillator reads According to the general formulation of section 2.2, for a given a functional F the time-ordered correlation function can be computed via a function f on R 2 that in cohomology is equivalent to More precisely, one determines a representative F ′ of F in the δ cohomology that can be written as . This means that F ′ should only depend on φ(t i ) and φ(t f ).
Since we want to compute the two-point function, we consider the functional that, for fixed t, s ∈ R, is defined by In order for F ′ = f • p to be in the same cohomology as F there should be a G such that F − F ′ = δG, where G has degree minus one.The perturbation lemma in the form (3.37) immediately gives us the function f as follows: where C is defined in terms of the Green's function (2.60) with Dirichlet boundary conditions: Note that here P 2 = P 1 since we are considering the free theory.We thus have using that the higher-order terms in vanish when acting on the functional (4.3) with two φ.
For the second term on the right-hand side we need to use (4.5) to compute for which one uses that K DD is symmetric.To evaluate then f (x, y) the first term in (4.6) maps x, y via the inclusion map (2.64) to a solution φ p with boundary conditions φ p (t i ) = x and φ p (t f ) = y and then evaluates the functional on this solution.Thus, we have Via the dictionary (4.2) this gives the desired two-point function.

Homological Computation for Coherent States
In the previous subsection we used the propagator with Dirichlet boundary conditions to find a certain representative in cohomology of the functional F = φ(t)φ(s).We want to see what happens when we instead use different propagators.The standard propagator in quantum field theory is the Feynman propagator, which we will investigate here.
In (1 + 0)-dimensions the Feynman propagator is given by Note that h F (f ) is complex, even when f is real.It is therefore not sufficient to work with the field space V , but rather we should work with the complexified field space V ⊗ C. We now want to derive the associated inclusion i and projection p F from the homotopy relation On the other hand, on equations of motion φ + ω 2 φ we find and so for the sum We next define new (complex) functionals a(t) and a † (t) by These expressions are motivated by the mode expansion of the harmonic oscillator, but we should emphasize that here these are just regular functions, not quantum operators.In particular, the function a † is just the complex conjugate of the function a, with the notation just reminding us of the usual raising and lowering operators.In terms of these we have This suggests that we define a projector p and the inclusion i F : with zero image in V 1 .With these definitions we have p For a given a field φ the projector p F gives the complex values a(t i ) and a † (t f ).Let us compare this with the projector associated to the propagator with Dirichlet boundary conditions, which gives (φ(t i ), φ(t f )).When relating to canonical quantization, we found that the correlator computed with the latter prescription used the states |x and y| satisfying φ(t i )|x = x|x and y|φ(t f ) = y|y.Correspondingly, when finding a representative F ′ of the cohomology of some functional F with F ′ = f • p F , we expect that f computes correlators with the in-state |z being an eigenstate of a(t i ) and the out-state w| being an eigenstate of a † (t f ).This will be confirmed in the following.
The eigenstates |z of the annihilation operator a are called coherent states.We use the convention a|z = z|z and z|a † = z|z.With this convention, z| is the conjugate of |z , where z denotes the complex conjugate of z.We can check whether it is reasonable that p F gives rise to correlators with coherent states by looking again at the two-point function.By either applying the perturbation lemma or going through the same steps as in the previous section, we find that the representative F ′ of the cohomology of We therefore claim that where It is straightforward to verify equation (4.18) in the familiar operator language of quantum mechanics, as we do now.We first recall that Wick's theorem implies where N is the normal ordering operation.We have the operator relation where a and a † are now interpreted as the creation and annihilation operators of the harmonic oscillator, satisfying the familiar commutation relations.Usually the above expression appears in textbooks for t i = t f = 0 and normal ordering is defined with respect to a := a(0) and a † := a † (0).But this is the same as normal ordering a(t i ) and a † (t f ), since they only differ from a and a † by phases.We can now compute by evaluating a(t i ) at z and a † (t f ) at w.This follows because N moves all annihilation operators to the right, where we can then use a(t i )|z = z|z .Similarly, when creation operators are on the right, we can use w|a † (t f ) = w|w.Therefore, Combining this with (4.20) then proves (4.18).
In case of the Feynman propagator, this result explains why the perturbation lemma gives Wick's theorem via the projector P ′ and P can be interpreted as normal ordering.Recall that P = i * just evaluates functionals on-shell, with boundary conditions specified by the inclusion map i.But this is just what we did in (4.23).We evaluated φ(t)φ(s) on the solution with a(t) = z at t = t i and a † (t) = w at t = t f .Of course, there is nothing special about the twopoint functions considered here, and so the perturbation lemma says that P ′ is really Wick's theorem squeezed between coherent states.

Comparison with Operator Language
In the previous two sections we applied the homological recipe to compute correlators with respect to position eigenstates and with respect to coherent states.The respective projectors were given by p(φ, φ Using Wick's theorem, for p F it was straightforward to see that our approach agrees with the operator language.For p, however, it is harder to verify that f defined via F ′ = f • p actually computes the correlator with respect to position eigenstates, although our formal path integral manipulations above suggest that this must be so.The general claim following from the homological approach is where f is the function whose pullback p * (f ) equals F = φ(t)φ(s) in cohomology.Our goal in this subsection is to check this statement using the standard formalism of quantum mechanics.
Since the operator computation is quite involved, for simplicity we set x = y = 0. Since the perturbation lemma immediately gives the full result (4.8) we see that the homological approach is advantageous in this case.When x = y = 0, (4.8) equals the Green's function K DD (t, s).So we want to establish the identity in the operator language.
We first review some more facts about coherent states.As already stated, a coherent state |z is an eigenstate of the annihilation operator, i.e. â|z = z|z for all z ∈ C . (4.27) We use the convention that the hermitian conjugate of |z is z|, so that z|â † = z|z.Given a general state |ψ , its overlap with a coherent state z| gives a holomorphic function in z, ψ(z) := z|ψ .(4.28) The inner product of two such states is The Hilbert space equipped with this inner product is called the Segal-Bargmann space.The identity can be written as The creation operator acts by multiplication since We now come back to the original goal of this section, i.e. establishing the identity (4.26) in the operator language.To do so, we use Wick's theorem This formula can be reduced to an integral over the reals.For example, one can show that [29] where C is some constant and x + iy| is a coherent state, and so x| = Ce −x 2 /2 dye −y 2 /2 x + iy| .We now use this in (4.36) to compute g(t, s).We first compute the denominator where we used the time evolution operator with respect to the Hamiltonian H = ω(a † a + 1 2 ) of the harmonic oscillator.Thus, using (4.40), we will need the time evolution of a coherent state.Defining T = t f − t i , we need to compute e i H T |z , which can be done by inserting a complete set of eigenstates |n of the Hamiltonian and using the overlap (4.33).With this, we find e i H T |z = e i ω 2 T |e −iωT z .Defining λ := e −iωT we thus have e i H T |z = λ − 1 2 |λz .Using this together with (4.40) we have: where we performed the Gaussian integral, and C 1 := 2πC 2 is another constant that will cancel in the end.Next we turn to the numerator of (4.36).Expanding φ(t) in terms of ladder operators, and using this in (4.36), we need to compute expectation values of operators quadratic in a and a † .For example, we find that (4.44) Similarly, we have Since the operators are normal ordered we do not need to compute y = 0; t f |aa † |x = 0; t i .We can now use the above to compute the normal ordered correlator in (4.36), for which we find after some algebra In order to relate this to K DD we rewrite the Feynman propagator, where K R (t, s) = θ(t − s) sin ω(t−s) 2ω is the retarded propagator.This yields where we reintroduced t i and t f through λ = e −iω(t f −t i ) .We can now make use of the identity cos ω((t to arrive at which is what we wanted to show.

General Boundary Conditions
In the previous subsection we exemplified our approach using two different projectors, which where given by p DD (φ) = (φ(t i ), φ(t f )) and p F (φ) = (a(t i ), a † (t f )).Our computations showed that these determine different types of correlation functions.
We now want to generalize to arbitrary linear boundary conditions.More precisely, we look at boundary conditions of the form where the numbers (a, b, c, d, x, y) can in general be complex.The numbers (x, y) parametrize solutions to the equations of motion.In this way, we obtain a projector This solution defines an inclusion which we extend to V • via the inclusion V 0 ֒→ V • .To find the homotopy h from the identity to i • p, we note that the homotopies h DD and h F satisfy the boundary conditions (4.52) with x = y = 0.So our ansatz for the homotopy h(f ) is the unique solution to φ+ ω 2 φ = f satisfying p • h = 0.It is given by where In kernel notation, we have Note that K(t, s) is manifestly symmetric in its arguments.A lengthy computation now shows that as well as ḧ(f ) + ω 2 h(f ) = f .Therefore, the homotopy relation {∂, h} = 1 − i • p is satisfied.
One application using these more general projectors and homotopies would be the computation of correlators with in-and out states living in different representations of the Hilbert space.For example, one could choose (a, b) = (1, 0) and (c, d) = (0, 1).In this case, the homotopy satisfies Dirichlet boundary conditions at t = t i and Neumann boundary conditions at t = t f .The associated representative of the cohomology then uses position eigenstates |x; t i as in-states and momentum eigenstates p; t f | as out-states.

Unruh Effect
In this section we present the first application of the homological formulation in the realm of genuine field theories.Specifically, we apply our homological method in the context of quantum field theory on curved spacetime by providing an alternative derivation of the Unruh effect: the quantum effect according to which the number of particles detected depends on the observer [30].In the vacuum state an inertial observer in Minkowski space sees no particles, while in the same state a uniformly accelerated observer sees a thermal bath of particles.

Generalities and Homotopy Retract
Let us begin with a brief review of general features of uniformly accelerated observers in twodimensional Minkowski spacetime with metric ds 2 = dt 2 − dx 2 . (5.1) The trajectory of an observer is then parametrized by x µ (τ ) = t(τ ), x(τ ) , where τ is proper time, so that the 2-velocity u µ (τ ) = dx(τ )/dτ satisfies the normalization condition η µν u µ u ν = 1 . (5. 2) The Lorentz-invariant condition for the acceleration being constant is expressed in terms of a µ (τ ) = uµ (τ ) as where a is a constant.The trajectory of a uniformly accelerated observer satisfying these two conditions can be written as Next, let us relate the inertial frame to a frame that is comoving with the observer.This means that denoting these coordinates by ( t, x) the observer's worldline is a vertical line x = 0, so that the observer is indeed at rest in this frame.The Rindler coordinates having this property are defined by t = a −1 e ax sinh a t , (5.5) and the inverse relation x = 1 2a ln a 2 (x 2 − t 2 ) . (5.8) From these relations one finds the metric in Rindler coordinates, which is thus conformally equivalent to the Minkowski metric.
We now consider the action of a massless scalar field φ in a 1 + 1 dimensional spacetime, where g µν is the metric and g is its determinant.In the inertial frame, (5.11) The action in the accelerated frame takes the same form: as a consequence of the conformal invariance of the action (5.10) in two dimensions and the Rindler metric (5.9) being conformally equivalent to the Minkowski metric.The equations of motion are φ − ∂ 2 x φ = 0 , (5.13) where the dot denotes the partial derivative with respect to time t.Note that as a scalar we have for the coordinate-transformed field φ( t, x) = φ(t, x), so that in the second equation we could replace φ by φ.
As a preparation for the homotopy retract we have to introduce the Fourier transform with respect to the spatial coordinate, both in inertial and Rindler coordinates: Note that even though in the second integral we could replace φ( t, x) by φ(t, x), the Fourier mode φ k as a function of k of course differs from φl as a function of l.The inverse relations are Since the scalar functions on the left-hand sides are equal (more precisely, we have φ(t, x) = φ( t(t, x), x(t, x))), we have two different expansions of the same φ into Fourier modes: φl t(t, x) . (5.17) We will also use the following change of basis for the Fourier modes and their time derivatives: where The inverse relations read: As for the harmonic oscillator these relations are motivated by the familiar definition of creation and annihilation operators, but we emphasize that also here these are just functions.
We now discuss the homotopy retract, beginning with the chain complex defining the theory: Here the space of fields and the space of anti-fields are given by The notation indicates that the (anti-)fields depend on t, restricted to the interval [t i , t f ], and the space coordinate x living on the full real line R.The differential is The important new feature in field theory is that the projector p : longer maps to a finite-dimensional space like R 2 but to infinite-dimensional functions spaces, however, with functions of one less coordinate.Specifically, the projector evaluates the functions a and a † defined in (5.20) at t i and t f , respectively: Next, we need to define the inclusion map i : C ∞ (R) × C ∞ (R) → V 0 that takes two functions in momentum space, say c(k) and d(k), and produces a field in V 0 (i.e. in the present example a scalar field in two-dimensional Minkowski space).The proper inclusion map satisfying p (5.26) The homotopy map h : where the kernel is explicitly given by Indeed, one can verify that with the above definitions for projector, inclusion and homotopy the homotopy relation ∂h + h∂ = 1 − ip is satisfied.To this end one needs to assume that φ(t, x) and ∂ x φ(t, x) vanish at x = −∞ and x = +∞.
For completeness we also display the important operations of the dual space of functionals on which the BV algebra is defined.The BV complex F(V • ) is equipped with the differential, In addition, the BV-differential is defined as (5.31) Similarly, the pullback of a functional (5.32)

Number Expectation Value
To derive the Unruh effect, one assumes that the number of particles measured by an accelerated observer is given by the expectation value of the number operator with respect to Rindler space, i.e., with respect to creation and annihilation operators defined with the Fourier modes in Rindler space.More precisely, one computes where bk and b † k are the Rindler space annihilation and creation operators defined in analogy to (5.21), and |0 is the Minkowski vacuum state.This state is defined so that it is annihilated by the inertial frame operator âk : âk |0 = 0 . (5.34) For definiteness we take the Heisenberg picture operators b and b † to be at time t = 0 (which is equivalent to t = 0 for all x).The usual textbook computation involves relating the creation and annihilation operators of the accelerated and inertial frames through Bogolyubov transformations.We provide an alternative approach which does not require finding the Bogolyubov transformations.Instead, our strategy is to define the functional F [φ] of the massless scalar field φ to be given by b † k b k , with b † k and b k being defined in terms of the classical field φ via (5.21).Following our approach for the harmonic oscillator in sec.and analogously for |d .Here we take the limit t → 0 after performing the computation, as opposed to setting t = 0 from the beginning, since some care is needed in order to deal with the step functions entering the Green's function.Note that the result does not depend on whether one takes the limit from above or from below, which follows from the symmetry of the Green's function.Finally, in order to find the expectation value of the Rindler number operator with respect to the Minkowski vacuum, we set c = d = 0, i.e., N k = f (0, 0) . (5.37) The choice c = d = 0 is the analog of the equation (5.34) specifying the Minkowski vacuum.
We begin by expressing the functional b † k ( t)b k (0) in terms of φ(t, x).By taking the Fourier transform of (5.21) where of course t and x on the right-hand side must be viewed as functions of ( t, x).
We apply the perturbation lemma to find f (c, d), by using P 1 in (3.34), (5.44) There are no further terms in the expansion of exp − i 2 C because F [φ] only contains two φs.Let us start by treating the first term on the right-hand side of (5.44).Since we set c = d = 0, the inclusion to the space of fields (5.26) is i(0, 0) = 0. Therefore, with (5.31), the first term on the right-hand side of (5.44) vanishes: i * F (0, 0) = 0 .
(5.45) Next, we take the limit t = t = 0.After inserting the derivatives of K(t, s), using (5.28), and writing these in terms of Rindler coordinates, we obtain f (0, 0) = P 1 (F )(0, 0) = dx dỹ dl Ω k 16π 2 ω l e ik(x−ỹ) e ila −1 (e ax −e a ỹ) (5.46) phase space, thereby circumventing the problem to make sense of the path integral over the full infinite-dimensional space.We have shown with a number of examples that the homological formulation allows one to compute concrete quantum expectation values that agree with those determined by standard quantum mechanics.However, so far this reformulation is not complete: it only provides a prescription to compute certain normalized quantum expectation values with respect to certain states.It then remains as the most important outstanding problem to explore whether this homological formulation could be completed to a full-fledged reformulation of quantum mechanics (and quantum field theory).
It is instructive to compare the techniques presented here with other approaches in the literature, see [22][23][24][25].The idea in these references is to pass via homotopy transfer from the L ∞ -algebra of a given theory to a 'minimal model' or 'on-shell' L ∞ -algebra on the cohomology (i.e. with all differentials being trivial).These L ∞ brackets compute (at least tree-level) scattering amplitudes, but one has to overcome some technical challenges.First, in order for the action and inner product to be well-defined the space of functions is restricted to Schwartz functions, but then there is no cohomology, no on-shell fields and hence no minimal model (for Schwartz functions φ = 0 implies φ = 0).One attempts to circumvent this problem by adding on-shell states by hand in degree zero and degree one, so that there is a minimal model.However, a priori the L ∞ brackets are then not well-defined since the product of two on-shell fields is neither on-shell nor a Schwartz function.In order to remedy this ref.[23] introduces certain regularizing factors in the products of fields.
In the homological formulation of this paper these issues do not arise.An important reason is that we do not consider smooth functions on R but rather on the finite integral [t i , t f ].Then it is not necessary to restrict to Schwartz functions and so there is non-trivial cohomology. 8owever, since in our formulation the cohomology is concentrated in degree zero there is no non-trivial L ∞ -algebra on this space; rather, the quantum expectation values are computed by the functions of the homotopy retract of the BV algebra, as described in the main text.Technically, the price to pay for working with general smooth functions on [t i , t f ] is that the symplectic form encoded in the anti-bracket is no longer invariant under the vector field Q.However, in our formulation the symplectic form does not enter, and so this does not cause any problems.(See also [19,20] where such a more general BV formalism was developed.) We close this paper with a brief list of interesting open problems: • Most intriguingly, the homological formulation is arguably mathematically well-defined for non-perturbative problems.It would then be important to apply and illustrate this method for genuinely non-perturbative problems.
• In this paper we have dealt with theories without gauge symmetries, so it would be interesting to consider gauge field theories such as Yang-Mills theory.Since the BV formalism was originally introduced in order to deal with subtle issues of gauge theories it should be straightforward to set up the corresponding BV algebra.However, it would still be instructive to work out the homotopy retract and the details of the homological formulation.
• One of the potentially most fruitful applications of the homological formulation may arise for quantum field theory on curved spacetime, where traditional flat space techniques to quantization have often been awkward.Our investigation was in fact motivated by the desire to find a systematic recipe to obtain in cosmological perturbation theory the quantum correlation functions directly from the L ∞ -algebra or the dual BV algebra.
In [28] we gave an interpretation of the passing over to gauge invariant so-called Bardeen variables of cosmological perturbation theory in terms of homotopy transfer but it remains to give a similar procedure for the computation of cosmological correlation functions.

(4. 53 )
As usual, we extend p toV • ⊗ C by setting p| V 1 = 0. We recover p DD when (a, b) = (c, d) = (1, 0), while p F is given by (a, b) = (c, d) = ( ω 2 , i ω 2 ).A solution with boundary conditions (4.52) is given by φ 4, we then find f (c, d) such that F ′ = f • p is in the same cohomology class as F [φ]. Then f (c, d) computes the expectation value f (c, d) = lim t→0 d|T b † k ( t) bk (0) |c d|c ,(5.35)where |c and |d are coherent states with respect to a k , i.e., a k |c = c(k)|c ,(5.36) This function f is unique since p * is invertible in cohomology.We claim that this function computes a certain normalized expectation value of F .More precisely, On the right-hand side, we use the language of canonical quantization, where T denotes the time ordering operation.Furthermore, the state |x; t is the state that satisfies the eigenvalue equation φ(t)|x; t = x|x; t , and similarly for y, where φ(t) is the canonically quantized position operator at time t associated to the classical field φ(t).
As a consistency check we note that [â, â † ] = 1 in this representation.Since the vacuum state |0 is annihilated by â, the vacuum is represented by ψ 0 (z) = z|0 that is in fact constant (and equal to one if we normalize it).Likewise, the nth excited state is , one obtains b k and b † k in terms of φ and ∂ tφ: b k ( t) = dx e −ikx Ω k ∂ x φ = e ax cosh(a t) φ + e ax sinh(a t)∂ x φ .(5.40)Note that this is only valid when x > |t| since the Rindler coordinates only cover this part of the Minkowski spacetime.With (5.38) -(5.40), we can explicitly write out the functional F [φ] = b † k ( t)b k (0):